LinkedIn·Tuesday, 25 August 2026·18h ago

Every wireless device transmits with small imperfections in its radio front-end. Those imperfections travel with the signal and make the…

Maarten Weyn

Every wireless device transmits with small imperfections in its radio front-end. Those imperfections travel with the signal and make the device recognisable, independently of its address, its key or its certificate. This is what we call a Radio Frequency Fingerprint. Deep learning separates those fingerprints well, as long as the devices are ones it was trained on, our models reach over 99 percent accuracy. In practice this means a network can recognise a device by its physical layer, without enrolling it first. For sensors too constrained for heavy cryptography, and for industrial networks where an unfamiliar transmitter needs to surface early, that is a usable direction. The same property has a flip side. A fingerprint that sits in the hardware itself does not disappear behind a randomised MAC address or behind encryption. What can identify a device can also track it. At IEEE WCNC 2026, Thayheng Nhem takes on that problem, and presented it in a per from himself together with Michael Peeters, Raf Berkvens and myself. Five unseen devices in a lab setup remains a first step, and the next one is a larger and more diverse test set. To be continued! Research at IDLab (UAntwerp - imec), University of Antwerp and imec supported by Research Foundation Flanders - FWO through the PESSO project. https://lnkd.in/eaSbYcee